Re:I worked on the US drone program. The public should know ...
by im3w1l on Sunday December 29, 2013 @09:41PM
(#6982721)
>I can point you to some fancy research in computer vision, that allows you to do just that.
Please do
Re:I worked on the US drone program. The public should know ...
by dchichkov on Monday December 30, 2013 @05:05PM
(#6987334)
Oh. I've tried finding that exact publication - they took a video of a city scene with dense fog/clouds in the background. If you'd play the video you couldn't see any shapes through the fog. At all. But after passing the whole video through reconstruction/de-noising, they've been able to recover the background (you could see s mountain, instead of dense fog).
It is not surprising and it's a pretty standard result actually. Can be called by different names (sparse sensing, denoising, superresolution, ...) and done by different methods (BP, LASSO, ...), but it always comes down to having a model (with some priors, e.g sparsity prior!) and using that model to re-project the data from your measurement projection into the projection of some interest to you.
I have no idea at what level you are, so I'd suggest this very well presented lecture (1 hour total): Compressed Sensing by Terence Tao http://www.youtube.com/playlist?list=PLC94A02A1218B24DF if you want to learn more. Examples of image recovery are in the part 5, part 6.
In case of multiple drones taking pictures, these can be synchronized using GPS time, down to a microsecond. And knowing the GPS coordinates of each drone you can recover the image. Pretty much with arbitrary quality. (& If you are really good at it and the object is stationary, you can do it even through heavy fog. I'm not kidding, you really can - I can point you to some fancy research in computer vision, that allows you to do just that.).